Vision Transformer-Based Tea Leaf Disease Classification Using a Novel Dataset

Tea is one of the most important cash crops in Bangladesh, contributing significantly to the national economy. However, its cultivation is frequently threatened by various leaf diseases that affect yield and quality. In this study, we introduce a novel, expert-annotated image dataset consisting of 3,300 tea leaf images—spanning six major diseases and healthy samples—collected from key tea-producing regions of Bangladesh under natural field conditions. The dataset is designed to facilitate the development and evaluation of automated disease detection systems using deep learning techniques. Each image has been meticulously labeled by tea experts based on field diagnosis and visual symptoms, ensuring biological and agricultural validity. Our work addresses domain-specific challenges such as class imbalance, inter-disease visual similarity, and noisy backgrounds. The dataset is partitioned into training, validation, and test sets (80: 10: 10) to ensure effective model development. This dataset serves as a foundational resource for researchers aiming to build scalable, AI-driven plant health monitoring systems tailored to the tea industry in developing regions.